1 / 23

Objective

ECE408 / CS483 Applied Parallel Programming Lecture 22: Application Case Study – Advanced MRI Reconstruction. Objective. To learn how to apply parallel programming techniques to an application Determining parallelism structure Loop transformations Memory layout considerations Validation.

rinah-jones
Download Presentation

Objective

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript

  1. ECE408 / CS483Applied Parallel ProgrammingLecture 22: Application Case Study – Advanced MRI Reconstruction

  2. Objective • To learn how to apply parallel programming techniques to an application • Determining parallelism structure • Loop transformations • Memory layout considerations • Validation

  3. Non-Cartiesian MRI Scan Cartesian Scan Data Spiral Scan Data Gridding1 FFT LS (a) (b) (c) 2

  4. Non-Cartesian Scan So Courtesy of Keith Thulborn and Ian Atkinson, Center for MR Research, University of Illinois at Chicago

  5. An Iterative Solver Based Apparch to Image Reconstruction Compute FHF + λWHW Acquire Data Compute FHD Find ρ

  6. Computation of Q and FHD for (m = 0; m < M; m++) { phiMag[m] = rPhi[m]*rPhi[m] + iPhi[m]*iPhi[m]; for (n = 0; n < N; n++) { expQ = 2*PI*(kx[m]*x[n] + ky[m]*y[n] + kz[m]*z[n]); rQ[n] +=phiMag[m]*cos(expQ); iQ[n] +=phiMag[m]*sin(expQ); } } (a) Q computation for (m = 0; m < M; m++) { rMu[m] = rPhi[m]*rD[m] + iPhi[m]*iD[m]; iMu[m] = rPhi[m]*iD[m] – iPhi[m]*rD[m]; for (n = 0; n < N; n++) { expFhD = 2*PI*(kx[m]*x[n] + ky[m]*y[n] + kz[m]*z[n]); cArg = cos(expFhD); sArg = sin(expFhD); rFhD[n] += rMu[m]*cArg – iMu[m]*sArg; iFhD[n] += iMu[m]*cArg + rMu[m]*sArg; } } (b) FHD computation

  7. First version of the FHD kernel. __global__ void cmpFhD(float* rPhi, iPhi, rD, iD, kx, ky, kz, x, y, z, rMu, iMu, rFhD, iFhD, int N) { int m = blockIdx.x * FHD_THREADS_PER_BLOCK + threadIdx.x; rMu[m] = rPhi[m]*rD[m] + iPhi[m]*iD[m]; iMu[m] = rPhi[m]*iD[m] – iPhi[m]*rD[m]; for (n = 0; n < N; n++) { float expFhD = 2*PI*(kx[m]*x[n] + ky[m]*y[n] + kz[m]*z[n]); float cArg = cos(expFhD); float sArg = sin(expFhD); rFhD[n] += rMu[m]*cArg – iMu[m]*sArg; iFhD[n] += iMu[m]*cArg + rMu[m]*sArg; } } 6

  8. Loop Fission for (m = 0; m < M; m++) { rMu[m] = rPhi[m]*rD[m] + iPhi[m]*iD[m]; iMu[m] = rPhi[m]*iD[m] – iPhi[m]*rD[m]; for (n = 0; n < N; n++) { expFhD = 2*PI*(kx[m]*x[n] + ky[m]*y[n] + kz[m]*z[n]); cArg = cos(expFhD); sArg = sin(expFhD); rFhD[n] += rMu[m]*cArg – iMu[m]*sArg; iFhD[n] += iMu[m]*cArg + rMu[m]*sArg; } } (a) FHD computation for (m = 0; m < M; m++) { rMu[m] = rPhi[m]*rD[m] + iPhi[m]*iD[m]; iMu[m] = rPhi[m]*iD[m] – iPhi[m]*rD[m]; } for (m = 0; m < M; m++) { for (n = 0; n < N; n++) { expFhD = 2*PI*(kx[m]*x[n] + ky[m]*y[n] + kz[m]*z[n]); cArg = cos(expFhD); sArg = sin(expFhD); rFhD[n] += rMu[m]*cArg – iMu[m]*sArg; iFhD[n] += iMu[m]*cArg + rMu[m]*sArg; } } (b) after loop fission

  9. cmpMu kernel __global__ void cmpMu(float* rPhi, iPhi, rD, iD, rMu, iMu) { int m = blockIdx.x * MU_THREAEDS_PER_BLOCK + threadIdx.x; rMu[m] = rPhi[m]*rD[m] + iPhi[m]*iD[m]; iMu[m] = rPhi[m]*iD[m] – iPhi[m]*rD[m]; }

  10. Second option of the FHD kernel __global__ void cmpFhD(float* rPhi, iPhi, phiMag, kx, ky, kz, x, y, z, rMu, iMu, int N) { int m = blockIdx.x * FHD_THREADS_PER_BLOCK + threadIdx.x; for (n = 0; n < N; n++) { float expFhD = 2*PI*(kx[m]*x[n]+ky[m]*y[n]+kz[m]*z[n]); float cArg = cos(expFhD); float sArg = sin(expFhD); rFhD[n] += rMu[m]*cArg – iMu[m]*sArg; iFhD[n] += iMu[m]*cArg + rMu[m]*sArg; } } 6

  11. Loop interchange of the FHD computation for (m = 0; m < M; m++) { for (n = 0; n < N; n++) { expFhD = 2*PI*(kx[m]*x[n] + ky[m]*y[n] + kz[m]*z[n]); cArg = cos(expFhD); sArg = sin(expFhD); rFhD[n] += rMu[m]*cArg – iMu[m]*sArg; iFhD[n] += iMu[m]*cArg + rMu[m]*sArg; } } (a) before loop interchange for (n = 0; n < N; n++) { for (m = 0; m < M; m++) { expFhD = 2*PI*(kx[m]*x[n] + ky[m]*y[n] + kz[m]*z[n]); cArg = cos(expFhD); sArg = sin(expFhD); rFhD[n] += rMu[m]*cArg – iMu[m]*sArg; iFhD[n] += iMu[m]*cArg + rMu[m]*sArg; } } (b) after loop interchange

  12. Third option of the FHD kernel __global__ void cmpFhD(float* rPhi, iPhi, phiMag, kx, ky, kz, x, y, z, rMu, iMu, int M) { int n = blockIdx.x * FHD_THREADS_PER_BLOCK + threadIdx.x; for (m = 0; m < M; m++) { float expFhD = 2*PI*(kx[m]*x[n]+ky[m]*y[n]+kz[m]*z[n]); float cArg = cos(expFhD); float sArg = sin(expFhD); rFhD[n] += rMu[m]*cArg – iMu[m]*sArg; iFhD[n] += iMu[m]*cArg + rMu[m]*sArg; } } 6

  13. Using registers to reduce memory accesses in the FHD kernel __global__ void cmpFhD(float* rPhi, iPhi, phiMag, kx, ky, kz, x, y, z, rMu, iMu, int M) { int n = blockIdx.x * FHD_THREADS_PER_BLOCK + threadIdx.x; float xn_r = x[n]; float yn_r = y[n]; float zn_r = z[n]; float rFhDn_r = rFhD[n]; float iFhDn_r = iFhD[n]; for (m = 0; m < M; m++) { float expFhD = 2*PI*(kx[m]*xn_r+ky[m]*yn_r+kz[m]*zn_r); float cArg = cos(expFhD); float sArg = sin(expFhD); rFhDn_r += rMu[m]*cArg – iMu[m]*sArg; iFhDn_r += iMu[m]*cArg + rMu[m]*sArg; } rFhD[n] = rFhD_r; iFhD[n] = iFhD_r; } 6

  14. Chunking k-space data to fit into constant memory __constant__ float kx_c[CHUNK_SIZE], ky_c[CHUNK_SIZE], kz_c[CHUNK_SIZE]; … __ void main() { inti; for (i = 0; i < M/CHUNK_SIZE; i++); cudaMemcpyToSymbol(kx_c,&kx[i*CHUNK_SIZE],4*CHUNK_SIZE, cudaMemCpyHostToDevice); cudaMemcpyToSymbol(ky_c,&ky[i*CHUNK_SIZE],4*CHUNK_SIZE, cudaMemCpyHostToDevice); cudaMemcpyToSymbol(ky_c,&ky[i*CHUNK_SIZE],4*CHUNK_SIZE, cudaMemCpyHostToDevice); … cmpFhD<<<FHD_THREADS_PER_BLOCK, N/FHD_THREADS_PER_BLOCK>>> (rPhi, iPhi, phiMag, x, y, z, rMu, iMu, CHUNK_SIZE); } /* Need to call kernel one more time if M is not */ /* perfect multiple of CHUNK SIZE */ } 6

  15. Revised FHD Kernel – Constant Memory __global__ void cmpFhD(float* rPhi, iPhi, phiMag, x, y, z, rMu, iMu, int M) { int n = blockIdx.x * FHD_THREADS_PER_BLOCK + threadIdx.x; float xn_r = x[n]; float yn_r = y[n]; float zn_r = z[n]; float rFhDn_r = rFhD[n]; float iFhDn_r = iFhD[n]; for (m = 0; m < M; m++) { float expFhD = 2*PI*(kx_c[m]*xn_r+ky_c[m]*yn_r+kz_c[m]*zn_r); float cArg = cos(expFhD); float sArg = sin(expFhD); rFhDn_r += rMu[m]*cArg – iMu[m]*sArg; iFhDn_r += iMu[m]*cArg + rMu[m]*sArg; } rFhD[n] = rFhD_r; iFhD[n] = iFhD_r; } 6

  16. Constant Memory Layout Consideration • Storing k-space samples in three separate arrays requires 3 cache lines for each warp • Interleaving x, y, z values of the same sample in the same array reduces the cache line requirement to 1 per warp (a) k-space data stored in separate arrays. (b) k-space data stored in an array whose elements are structs. 25

  17. Host Code with Adjusted Constant Memory Layout structkdata { float x, float y, float z; } k; __constant__ structkdatak_c[CHUNK_SIZE]; … __ void main() { inti; for (i = 0; i < M/CHUNK_SIZE; i++); cudaMemcpyToSymbol(k_c,k,12*CHUNK_SIZE, cudaMemCpyHostToDevice); cmpFhD<<<FHD_THREADS_PER_BLOCK, N/FHD_THREADS_PER_BLOCK>>> (…); } } 6

  18. Adjusted k-space data constant memory layout in the FHD kernel __global__ void cmpFhD(float* rPhi, iPhi, phiMag, x, y, z, rMu, iMu, int M) { int n = blockIdx.x * FHD_THREADS_PER_BLOCK + threadIdx.x; float xn_r = x[n]; float yn_r = y[n]; float zn_r = z[n]; float rFhDn_r = rFhD[n]; float iFhDn_r = iFhD[n]; for (m = 0; m < M; m++) { float expFhD = 2*PI*(k[m].x*xn_r+k[m].y*yn_r+k[m].z*zn_r); float cArg = cos(expFhD); float sArg = sin(expFhD); rFhDn_r += rMu[m]*cArg – iMu[m]*sArg; iFhDn_r += iMu[m]*cArg + rMu[m]*sArg; } rFhD[n] = rFhD_r; iFhD[n] = iFhD_r; } 6

  19. Using Hardware __sin() and __cos() __global__ void cmpFhD(float* rPhi, iPhi, phiMag, x, y, z, rMu, iMu, int M) { int n = blockIdx.x * FHD_THREADS_PER_BLOCK + threadIdx.x; float xn_r = x[n]; float yn_r = y[n]; float zn_r = z[n]; float rFhDn_r = rFhD[n]; float iFhDn_r = iFhD[n]; for (m = 0; m < M; m++) { float expFhD = 2*PI*(k[m].x*xn_r+k[m].y*yn_r+k[m].z*zn_r); float cArg = __cos(expFhD); float sArg = __sin(expFhD); rFhDn_r += rMu[m]*cArg – iMu[m]*sArg; iFhDn_r += iMu[m]*cArg + rMu[m]*sArg; } rFhD[n] = rFhD_r; iFhD[n] = iFhD_r; } 6

  20. Validating Reconstructed Image Using Peak Signal-to-Noise Ratio A.N. Netravali and B.G. Haskell, Digital Pictures: Representation, Compression, and Standards (2nd Ed), Plenum Press, New York, NY (1995). I0 is a known, “perfect” answer. This is typically done by creating k-space samples for a known image, producing a reconstructed image, and compare the two. 20

  21. Validation of floating-point precision and accuracy of the different FHD implementations. Is the known image answer (sometimes called a phantom image) Note that all GPU optimized versions have comparable PSNR as (3) the CPU double-precision version

  22. Component and Whole-Application Speedup 108X 357X 228X 30

  23. Any Questions?

More Related